ADC-Less 3D-NAND Compute-in-Memory Architecture Using Margin Propagation

Compute In Memory (CIM) has gained significant attention in recent years due to its potential to overcome the memory bottleneck in Von-Neumann computing architectures. While most CIM architectures use non-volatile memory elements in a NOR-based configuration, NAND-based configuration, and in particular, 3D-NAND flash memories are attractive because of their potential in achieving ultra-high memory density and ultra-low cost per bit storage. Unfortunately, the standard multiply-and-accumulate (MAC) CIM-paradigm can not be directly applied to NAND-flash memories. In this paper, we report a NAND-Flash-based CIM architecture by combining the conventional 3D-NAND flash with a Margin-Propagation (MP) based approximate computing technique. We show its application for implementing matrix-vector multipliers (MVMs) that do not require analog-to-digital converters (ADCs) for read-out. Using simulation results we show that this approach has the potential to provide a 100 x improvement in compute density, read speed, and computation efficiency compared to the current state-of-the-art.

内存计算（CIM）近年来因其有可能克服冯·诺依曼计算架构中的内存瓶颈而受到广泛关注。虽然大多数CIM架构在基于NOR的配置、基于NAND的配置中使用非易失性存储元件，特别是3D - NAND闪存因其在实现超高存储密度和每比特存储超低成本方面的潜力而具有吸引力。不幸的是，标准的乘法累加（MAC）CIM范式无法直接应用于NAND闪存。在本文中，我们通过将传统的3D - NAND闪存与基于边际传播（MP）的近似计算技术相结合，报告了一种基于NAND闪存的CIM架构。我们展示了其在实现矩阵 - 向量乘法器（MVM）方面的应用，该乘法器在读取时不需要模数转换器（ADC）。通过模拟结果我们表明，与当前最先进的技术相比，这种方法有可能在计算密度、读取速度和计算效率方面提供100倍的提升。